library(mclust)
## Package 'mclust' version 6.1.1
## Type 'citation("mclust")' for citing this R package in publications.
library(ggplot2)

# 1. Leer datos
wf_concat3 <- read.csv("wf_concat3.csv")
wf_concat3 <- as.matrix(wf_concat3)
dim(wf_concat3)
## [1] 38245   180
# Seleccionar la fila que querés visualizar
i <- 889
wave <- wf_concat3[i, ]
n <- length(wave)

# Cantidad de samples por canal
samples_per_channel <- 45

# Calcular posiciones de separación
separadores <- seq(samples_per_channel, n, by = samples_per_channel)

# Graficar
plot(wave, type = "l", col = "blue",
     main = paste("Waveform concatenada - spike", i),
     xlab = "Muestras concatenadas",
     ylab = "Amplitud")

# Agregar líneas verticales en las fronteras de cada canal
abline(v = separadores, col = "red", lty = 2)

library(ks)
library(ggplot2)

pca <- prcomp(wf_concat3, center = TRUE)
X <- pca$x[, 1:2]   # TOMAR SOLO 2 COMPONENTES

x <- X              # Renombrar para claridad
print(dim(x))       # Debe ser Nx2
## [1] 38245     2
H <- Hpi(x)         # Bandwidth KDE
k <- kde(x, H = H)  # KDE 2D

df <- expand.grid(
  x = k$eval.points[[1]],
  y = k$eval.points[[2]]
)

df$z <- as.vector(k$estimate)

ggplot(df, aes(x, y, fill = z)) +
  geom_tile() +
  scale_fill_viridis_c() +
  theme_minimal() +
  labs(
    title = "Densidad KDE en espacio PCA",
    x = "PC1",
    y = "PC2",
    fill = "Densidad"
)

# 2. PCA 
pca <- prcomp(wf_concat3, center = TRUE, scale. = FALSE)
X <- pca$x[, 1:4]   # Usar 4 componentes como en Python

# 3. Ajustar GMM con EM
# Aquí defines el número de clusters:
k <- 5

gmm <- Mclust(X, G = k, modelNames = "VVV")  
# "VVV" = covarianza full, equivalente a covariance_type="full"

labels <- gmm$classification

# 4. Graficar PC1 vs PC2 coloreado por cluster
df_plot <- data.frame(PC1 = X[,1], PC2 = X[,2], cluster = as.factor(labels))

ggplot(df_plot, aes(x = PC1, y = PC2, color = cluster)) +
    geom_point(size = 1, alpha = 0.3) +
    scale_color_brewer(palette = "Dark2") +
    theme_minimal() +
    ggtitle("Clusters detectados por GMM (EM)") +
    theme(legend.position = "right")

library(ggplot2)
k <- 5
# K-means directamente sobre los datos originales
labels_kmeans <- kmeans(wf_concat3, centers = k, nstart = 20)$cluster

# PCA para graficar
pca2 <- prcomp(wf_concat3, center = TRUE)
X2 <- pca2$x[, 1:2]

df_kmeans <- data.frame(PC1 = X2[,1], PC2 = X2[,2],
                        cluster = as.factor(labels_kmeans))

ggplot(df_kmeans, aes(x = PC1, y = PC2, color = cluster)) +
    geom_point(size = 1, alpha = 0.3) +
    scale_color_brewer(palette = "Dark2") +
    theme_minimal() +
    ggtitle("Clusters K-means (K=4) en espacio PCA")

library(mclust)
library(ggplot2)

# 1. Leer datos
wf_concat2 <- read.csv("wf_concat2.csv")
wf_concat2 <- as.matrix(wf_concat2)
dim(wf_concat2)
## [1] 52557   180
# Seleccionar la fila que querés visualizar
i <- 1
wave <- wf_concat2[i, ]
n <- length(wave)

# Cantidad de samples por canal
samples_per_channel <- 45

# Calcular posiciones de separación
separadores <- seq(samples_per_channel, n, by = samples_per_channel)

# Graficar
plot(wave, type = "l", col = "blue",
     main = paste("Waveform concatenada - fila", i),
     xlab = "Muestras concatenadas",
     ylab = "Amplitud")

# Agregar líneas verticales en las fronteras de cada canal
abline(v = separadores, col = "red", lty = 2)

library(ks)
library(ggplot2)

pca <- prcomp(wf_concat2, center = TRUE)
X <- pca$x[, 1:2]   # TOMAR SOLO 2 COMPONENTES

x <- X              # Renombrar para claridad
print(dim(x))       # Debe ser Nx2
## [1] 52557     2
H <- Hpi(x)         # Bandwidth KDE
k <- kde(x, H = H)  # KDE 2D

df <- expand.grid(
  x = k$eval.points[[1]],
  y = k$eval.points[[2]]
)

df$z <- as.vector(k$estimate)

ggplot(df, aes(x, y, fill = z)) +
  geom_tile() +
  scale_fill_viridis_c() +
  theme_minimal() +
  labs(
    title = "Densidad KDE en espacio PCA",
    x = "PC1",
    y = "PC2",
    fill = "Densidad"
)

# 2. PCA 
pca <- prcomp(wf_concat2, center = TRUE, scale. = FALSE)
X <- pca$x[, 1:4]   # Usar 4 componentes como en Python

# 3. Ajustar GMM con EM
# Aquí tú defines el número de clusters:
k <- 3

gmm <- Mclust(X, G = k, modelNames = "VVV")  
# "VVV" = covarianza full, equivalente a covariance_type="full"

labels <- gmm$classification

# 4. Graficar PC1 vs PC2 coloreado por cluster
df_plot <- data.frame(PC1 = X[,1], PC2 = X[,2], cluster = as.factor(labels))

ggplot(df_plot, aes(x = PC1, y = PC2, color = cluster)) +
    geom_point(size = 1, alpha = 0.3) +
    scale_color_brewer(palette = "Dark2") +
    theme_minimal() +
    ggtitle("Clusters detectados por GMM (EM)") +
    theme(legend.position = "right")

library(ggplot2)

# K-means directamente sobre los datos originales
labels_kmeans <- kmeans(wf_concat2, centers = k, nstart = 20)$cluster

# PCA para graficar
pca2 <- prcomp(wf_concat2, center = TRUE)
X2 <- pca2$x[, 1:2]

df_kmeans <- data.frame(PC1 = X2[,1], PC2 = X2[,2],
                        cluster = as.factor(labels_kmeans))

ggplot(df_kmeans, aes(x = PC1, y = PC2, color = cluster)) +
    geom_point(size = 1, alpha = 0.3) +
    scale_color_brewer(palette = "Dark2") +
    theme_minimal() +
    ggtitle("Clusters K-means (K=4) en espacio PCA")

Criterio BIC

pca <- prcomp(wf_concat3, center = TRUE)
X <- pca$x[, 1:6]
X_plot <- pca$x[, 1:2]
library(mclust)

em_fit <- Mclust(X, G = 1:6)    # Prueba G = 1:9, varios modelos
summary(em_fit)
## ---------------------------------------------------- 
## Gaussian finite mixture model fitted by EM algorithm 
## ---------------------------------------------------- 
## 
## Mclust VVV (ellipsoidal, varying volume, shape, and orientation) model with 6
## components: 
## 
##  log-likelihood     n  df      BIC      ICL
##        -1508709 38245 167 -3019179 -3033706
## 
## Clustering table:
##    1    2    3    4    5    6 
## 4713 8824 9905 5313 6267 3223
plot(em_fit, what = "BIC")

labels_em <- em_fit$classification
k_em <- em_fit$G
k_em
## [1] 6
df_em <- data.frame(
  PC1 = X_plot[,1],
  PC2 = X_plot[,2],
  cluster = as.factor(labels_em)
)

ggplot(df_em, aes(PC1, PC2, color = cluster)) +
    geom_point(size = 1, alpha = 0.4) +
    theme_minimal() +
    scale_color_brewer(palette = "Set1") +
    ggtitle(paste("EM/GMM con K óptimo por BIC =", k_em))